From: Stuart Brooks <stuartb@cat.co.za>
To: pgsql-sql@postgresql.org
Subject: Am I wasting my time with partitions?
Date: Tue, 19 Feb 2008 12:54:14 +0200
Message-ID: <47BAB556.7060506@cat.co.za> (raw)
It seems to me that postgresql doesn't use indexes when being asked for
an ordered result sets from a partitioned table. I have an application
where this is critical, but I was hoping to use partitions because of
the ease of rotating out old rows.
Simply put, I have a table called LineItems which I need to be able to
page from and so I need to be able to ask for N rows ordered on a
certain index (with possible constraints).
eg. SELECT * FROM T ORDER BY col1,col2 LIMIT 10;
This works fine and is quick on a single table:
>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>
metadb=> \d lineitems
Table "test2.lineitems"
Column | Type | Modifiers
--------------+--------------------------------+-----------
lineitem_key | bigint | not null
time | timestamp(6) without time zone | not null
description | text | not null
barcode | text | not null
amount | bigint | not null
Indexes:
"lineitems_amount_index" btree (amount, lineitem_key)
metadb=> explain select * from lineitems order by amount,lineitem_key
limit 10;
QUERY PLAN
--------------------------------------------------------------------------------------------------------
Limit (cost=0.00..0.74 rows=10 width=49)
-> Index Scan using lineitems_amount_index on lineitems
(cost=0.00..39791.76 rows=535500 width=49)
(2 rows)
>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>
If I partition the table by creating a top level table L, and inherited
tables L1, L2 and issue the same request it does sequential scans on all
the tables and takes orders of magnitude longer (see below).
In the example below I would have hoped that it would have used an index
scan on each of the tables returning 10 rows each and then done a merge
on them. Am I asking too much? Should I just use a single table and take
the hits on deletes and vacuums?
Regards
Stuart
>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>
metadb=> \d L
Table "test2.l"
Column | Type | Modifiers
--------------+--------------------------------+-----------
lineitem_key | bigint | not null
time | timestamp(6) without time zone | not null
description | text | not null
barcode | text | not null
amount | bigint | not null
Indexes:
"l_amount_index" btree (amount, lineitem_key)
metadb=> \d L1
Table "test2.l1"
Column | Type | Modifiers
--------------+--------------------------------+-----------
lineitem_key | bigint | not null
time | timestamp(6) without time zone | not null
description | text | not null
barcode | text | not null
amount | bigint | not null
Indexes:
"l1_amount_index" btree (amount, lineitem_key)
Inherits: l
metadb=> \d L2
Table "test2.l2"
Column | Type | Modifiers
--------------+--------------------------------+-----------
lineitem_key | bigint | not null
time | timestamp(6) without time zone | not null
description | text | not null
barcode | text | not null
amount | bigint | not null
Indexes:
"l2_amount_index" btree (amount, lineitem_key)
Inherits: l
metadb=> explain select * from l order by amount,lineitem_key limit 10;
QUERY PLAN
-------------------------------------------------------------------------------------
Limit (cost=22207.70..22207.72 rows=10 width=88)
-> Sort (cost=22207.70..23548.09 rows=536156 width=88)
Sort Key: test2.l.amount, test2.l.lineitem_key
-> Result (cost=0.00..10621.56 rows=536156 width=88)
-> Append (cost=0.00..10621.56 rows=536156 width=88)
-> Seq Scan on l (cost=0.00..16.90 rows=690 width=88)
-> Seq Scan on l1 l (cost=0.00..4951.00
rows=250000 width=49)
-> Seq Scan on l2 l (cost=0.00..5653.66
rows=285466 width=49)
(8 rows)
NB. Just addressing one of the inherited tables works fine.
metadb=> explain select * from l1 order by amount,lineitem_key limit 10;
QUERY PLAN
------------------------------------------------------------------------------------------
Limit (cost=0.00..0.74 rows=10 width=49)
-> Index Scan using l1_amount_index on l1 (cost=0.00..18554.20
rows=250000 width=49)
(2 rows)
>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Reply to all the recipients using the --to and --cc options:
reply via email
To: pgsql-sql@postgresql.org
Cc: stuartb@cat.co.za
Subject: Re: Am I wasting my time with partitions?
In-Reply-To: <47BAB556.7060506@cat.co.za>
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
This inbox is served by DDX for PostgreSQL; see mirroring instructions
for how to clone and mirror all data and code used for this inbox